Learn how to build a personal agent for task management, calendar access, and feedback analysis. Discover a pipeline for meeting transcription, prompt engineering, and generating actionable insights for weekly improvement.
I have a personal agent setup which runs on a cloud VM and helps me through various tasks. It has access to my calendar, meeting recordings, interviews, feedback, daily notes, notes management, learning new skills, managing my talks, my podcast, my finance, etc. In this session I covered a very specific use case that I’m using around getting feedback for my one-on-one meetings. I showed them how I have set up my pipeline that records and transcribes my meetings and stores them in specific notes in my Obsidian, we reviewed the prompts and the rubrics that are used to analyze those transcripts and generate feedback. At the start of the week, that feedback is aggregated, and one insight is generated to show on my daily note. This allows me to work on that feedback throughout the week and then see if I’ve done better the following week when the results come in again.
There is no project URL for this since this is my personal setup on my local machine.
Video
Transcript
Generated about 1 month ago
Summary
Generating a talk summary...
View full transcript
Speaker 0: AI, everyone. Thank you for coming here. ISO, yeah, this is the introductory Event, an Inaugural event for, the AI Tinkerers community in Islamabad. So this is a big global Community, over a 100,000 people globally. Last I checked, there are about 2 31 cities.
Speaker 0: We have a chapter in Lahore. I think they had an event in, December maybe, But, this is the latest chapter linting, Pakistan, and hopefully, we hope to make it a monthly event. So we're looking forward to all of you attending, spreading the word, and building the community. And as you already know, this is, no PowerPoint, no fluff. We all have to be building covering.
Speaker 0: So that's the criteria, and we'll talk more in the networking portion about what each of us are building. So little bit more about AI Tinkerers. Lead Ch, it's a big community. It's a global community. It's full of builders, investors, researchers.
Speaker 0: So you can tap into the community yourself to partner with people. I was just looking at people who are looking for partners for some global hackathon. It has a $100,000 AI. So you have the opportunity note just to partner with each author, but also the global community to showcase your work also. So if you're building something, you're looking for customers, investors, partners, Founder/CEO.
Speaker 0: So there's a lot of opportunity here. Right? And as you know, there's a lot of hype around AI, but we are all about the subagents, no hype. So really excited to have all of you here. Thank you for being here.
Speaker 0: Just a word of thanks to, Ali, 1st of all. They have partly sponsored the event. You'll get some yummy food, sponsored by Poly in a little while. And to Ch, who will give an intro himself about, Ali Ahmed 2nd, what they do and what he does here. But, as you know, we had a last minute change in venue.
Speaker 0: So thank you to Mashhood for, really stepping up and offering us the Spike, also for being 1 of the, speakers, for the demo. And, I also want to thank Daftarkhwan because they also are 1 of the sponsors. They offered us their space. As a venue partner, we had to change some plans last minute, but they had everything set up. Hopefully, we can continue partnering with them also in the future.
Speaker 0: So that's pretty much it. Hopefully, all of you have access to the platform, AI Tinkerers. You have you're logged in. You have access to it. It's a platform that's still developing.
Speaker 0: So if you have any bugs, you face any issues there, let us note. And looking forward to all the demos and also talking to all of you after the talks. So the format will be we'll just do 3 demos, maybe 4 IFC, 1 of our science fair people wants to set up, step up, Ch, but we also have a possibly science fair portion after the demos for people to showcase things they are working on. And after that, it's just networking and, kind of, you know, pizzas and so on. So it's very Information.
Speaker 0: Nothing nothing for Team. So thank you, Mashhood. Over to you. It's alright. I think the screen may be captured by me.
Speaker 0: Alright. ISO let me connect this. Alright. Everyone. I am head of engineering at.
Speaker 0: So I've been here for about 3 years, and, IOT do we do? Raise your hands if you've heard about Taleemabad before. ISO we are, we call ourselves EdTech, and, I think it's interesting because EdTech is a very AI space. We don't realize it. Our focus here, that's on, teachers, more specifically on public schools REACH because we feel improvement that's definitely 1 space Sorry.
Speaker 0: We'll we'll connect that. So, so over the last 10 years, we've been trying to figure out what works in schools, because we have a lot of demonstrates. Expensive technology Most of the time, some schools are very remote. You also have to sell it in a way so that the government understands the value. They're always cost cutting.
Speaker 0: Right? So So our Code product that we've been working on over the Rastgar years, it's called the digital coach. The idea is when teacher training happens, you send someone into the school to to see how the teacher is teaching, and that person coaches the teacher. Digital form digital coach REACH classroom record AI we use the audio file to understand 2nd we're also listening to the engagement of the students and everything and feedback there. That's AEC Code product.
Speaker 0: There's IOT of other stuff AI lesson plans, exam generation, exam checking stuff. Your teachers could help attendance Solo so that's what we're doing. All of this is obviously heavily using AI. And we're always trying to see how do you push that boundary, not in Team of just building the product, but AI also as an Organization. How how are organizations using to change as we use more and more of these, things as Full?
Speaker 0: So so that's about. For the the testing I wanna show today so 1 of the things that I am we we started doing the early Jun was trying to see what are all the things that we can automate in our in our work. Not in our work, IOT personal work AEC our work Jun Munib general. So what you see over here is the code Date that I started off, almost a year ago. It's gonna be a Mars PA.
Speaker 0: This is essentially a tool that's supposed to help me, do all the different things that I'm doing in my personal life. It's because of zoom in, ISO you guys can see terminal is not important. This is by by the way, this is my setup. This is how my this is how I work. So I usually have a, like, a Versus Code covering over here to review the files, review the Code, and this is my terminal.
Speaker 0: This is my cloud. Cloud may you see there's multiple tabs ISO can know what's happening. Yeah. Usually cloud coded, so that I can see different different Stack, AI can have, like, a tab covering like this. Joe, you can see Head AI have, like, a bunch of tabs running for something, or it's gonna hide hide tab color coded That's my setup.
Speaker 0: Synthesis Date there's a lot of stuff that's happening in this project. Qasim Since I only wanna spend 5 minutes, I wanna focus on just 1 thing. So the coolest thing about AI that I love about is IOT can give you feedback. And I am a big proponent of growth. So I've been using this tool to improve myself specifically in 1 area.
Speaker 0: Head spent a lot of AI. That is my 1 on ones with my team members. So I'm gonna pick on that, and I'm gonna show you guys what I do. AI head of Engineering. Team it, with my team leads.
Speaker 0: Yeah. Sometimes I do, skip level Jun on ones. I'm sometimes moving across Team, even in product, digital learning in the 1 on Agents. So, I have this tool over here. Managing Event.
Speaker 0: Any meeting that I do in person or online, it gets recorded on my computer. And when it gets recorded, it will get picked up, it will get transcribed, and then it will get processed. Joe the rules IOT this particular case AI in, adjacent form. So this is, a configuration template. Joe can see different define patterns so skip recordings with Hania.
Speaker 0: So it knows it has all of the information over harness, and then securing. So I have different recordings. I have 1 on 1 meetings over here. I have I hope I'm not blocking the view. So I have recordings.
Speaker 0: I have AI have Team syncs. So I'm doing a lot of team syncs, engineering, product, 2nd then I have other types of meetings below that. So you say meetings generated 2nd then I can control case meetings evaluate using these prompts. ISO if I have to Intake 1 on 1 AI have 3 different perspectives that I'm looking TBA 1 on 1 at. IOT this Jun general summary.
Speaker 0: 2nd 1 is feedback AEC it positive, negative to that person. And the 3rd 1 is evaluate which is the interesting part for me. This second happen with ISO this is the prompt for the self evaluation. Now, because we're in AEC 2nd we do a lot of observation of teachers, so this is our domain. This is where we're very strong at.
Speaker 0: So AI that we're using for our teachers as well in terms of, like, how much agency am I giving my, the my team member? Ownership Ch the monkey on his back or my back? Evaluate LLM I showing up in a specific way? Shutdown points. Right?
Speaker 0: So this is a rubric. Transcript and it generates a feedback note for me. So that feedback note gets stored in my Obsidian. So these are all AI note. Access Axis.
Speaker 0: AI note have City notes folder 2nd AI I can see my feedback and AI to give you guys a glimpse of how this looks AI, I think there should have been a feedback recently automate. Let's pick on this 1. So I had a, I had 1, I think, yesterday. It's gonna it can you can see over here, it's securing me. So on AG, on agency and ownership, I just got a 2 out of AI.
Speaker 0: And it's telling 2026, AI the start of a AEC. 2nd, again, it's I'm learning from IOT. So I I am getting this. Intern of telling herself, I was asking her to help me understand but then when 2nd it's mark me down on that. It's like, no.
Speaker 0: Don't do that. Let her figure it out. Just push her in the right direction. That's how you should do it. ISO you can see all of these files that are being Jun.
Speaker 0: Lots of feedback. Now I will sometimes come in and read this, but note, this gets reflected in my so this is this whole system also generates this daily note for me. This may only say meetings or may to do list. It reads last week meetings possibly the strongest STEP 2nd it keeps it here for the whole week. So this 1 thing AI have to think about as I open this folder every time because this Skill be a 2026 do list STEP.
Speaker 0: So I'm reminded again and again testing So as an example, 1 thing that I've improved over the last several weeks lead was on the Announced of time I second talking. So 1 on 1 scanner, normally, you want the other person to spend about 60 to 70% of your time talking, but it was the opposite in my case where I don't know how, but I was spending 70% of time talking. So it will remind me over here when Jero starts TBA Code that, 6050% AI think last week, I reached about 45%, which was AI an achievement for me. So this is how I'm doing this. Ali of this is running on, BPS.
Speaker 0: So I've set up a Google AI. This is open Chrome, set up Keyaway, and it has access to this code BIM. It has access to my notes. So all of these things are running in the background. They don't actually run on my COMPLIANCE.
Speaker 0: So it generates the build note, all of these, summary here, your notes since I Ch. 2nd then there's this Dropbox folder that's linked between my computer and that Community. So I can then go 2nd review Jun if I need to. So note ISO in this case, this PASSED Code, we have weekly analysis Claude it's telling me AI Folder using. Team reflection create Head LLM AI Code ISO I also get to keep tabs.
Speaker 0: I AI, I AI can use this. We can talk more about yep. That was it on on my side. You wanna do question Announced or we alright. Any questions, thoughts on this?
Speaker 1: AEC.
Speaker 0: Yeah. Jun, OpenClaw subscription 2nd that is running. That's it. Nothing else. Codec key subscription he use for everything.
Speaker 0: Yes. So you can configure, so there are 2 ways of doing things. 1 way is, orchestrator June 2nd then orchestrator calls APIs. Ch ISO this is what we're using Head. Subscription Ch in, cloud now, if you use your subscription for cloud AI speed of Venue Code that's where it is.
Speaker 0: IOT Codex codecs Codex.
Speaker 1: ISO, the meetings that you have, the 1 on ones,
Speaker 0: are they mix of in person and virtual? Yes. Both. Both. How do you manage, like, the transcription?
Speaker 0: Both. Or on my computer. So may I have the computer record So do in person audio team or we also have those mics in our room AEC mic is very good. So it picks up very easily. AI complex Custom Intake up the audio 2nd it does both sides.
Speaker 1: And how does your team feel about this? Do they
Speaker 0: also have
Speaker 1: access
Speaker 0: to this? No. They, they don't have access to these notes, but Folder appraisers. So when we generate feedback at the end of the cycle, so that's when they get to see all of these notes feedback. So so that's how it works right now.
Speaker 0: Codecs open ISO what happens is STEP of calling the API and using an API key, I talk the subscription to do it. That's AI context storage, that's my Obsidian note. So whenever it needs context, then it goes into my notes folder and checks out the console context. Yes. Yes.
Speaker 0: That's how it's configured. It starts a new new session every time. Team don't know if you guys can see it, but so, this is an open floor project I'm trying to learn philosophy more because it's interesting to he's a project manager So there are specific channels. AI then I have a general channel. Intern terms of brain terms of IOT perspective are you taking?
Speaker 0: In terms
Speaker 1: of
Speaker 0: Yeah. Yeah. I mean, so, like, it's a virtual example AI, I mean, I can, but the value generated for me just reduces significantly in most meetings. For me, June team testing Opus Skill. I want to know how I'm showing up in the team meetings.
Speaker 0: Is not a big deal. But what I'm more interested in is what can I lead, what can I use later on, what is useful Context? You can filter Jun will have so much noise in your data 2nd cleaning that up and using that will be very difficult. So part of that is constantly thinking about what is useful and what is not useful, and only keeping the useful stuff. Yeah.
Speaker 0: AI. I think that's a great question. I think, so when I started off, I I put on my drafting Engineer hat or system or documents architecture event sourcing because it's a very important system. It's gonna be my it's gonna be doing everything for me. It has to take all of these considerations.
Speaker 0: But I quickly realized second, I feel this is still too much code that I have. I mean, it's got a Jun there. That is basically just a bunch of Skill files. So all of this that I'm sharing is basically now encode ISO, my intention was PASSED now most of it REACH handles itself. So my learning has been and it just does it AI.
Tech stack
LLM
Large Language Models (LLMs) are deep learning models, built on the Transformer architecture, that process and generate human-quality text and code at scale.
LLMs are a class of foundation models: massive, pre-trained neural networks (often with billions to trillions of parameters) that leverage the self-attention mechanism of the Transformer architecture (introduced in 2017) to predict the next token in a sequence. Trained on vast datasets (e.g., Common Crawl's 50 billion+ web pages), these models—like GPT-4, Gemini, and Claude—acquire predictive power over syntax and semantics. They function as general-purpose sequence models, enabling critical applications such as complex content generation, language translation, and automated code completion (e.g., GitHub Copilot). Their core value: generalizing across diverse tasks with minimal task-specific fine-tuning.
OpenClaw is the viral, open-source, autonomous AI agent: a self-hosted 'digital employee' that executes real-world tasks across your local machine and messaging platforms 24/7.
This is the next-generation autonomous AI agent, built by Peter Steinberger (founder of PSPDFKit). OpenClaw functions as a proactive, self-hosted assistant, running as a long-running Node.js service on your own hardware (e.g., a Mac Mini or VPS) for about $3–$5 per month. It integrates directly with chat apps (WhatsApp, Telegram, Discord) to receive instructions and report completions. The agent utilizes over 100 AgentSkills to execute complex, real-world workflows: clearing your inbox, writing code, managing documents, and checking you in for flights. The open-source project’s velocity is undeniable, having surpassed 100,000 GitHub stars quickly and reportedly driving a surge in Mac Mini sales.
Cloud VMs provide on-demand, scalable virtual computing instances that run isolated operating systems on shared physical hardware.
Cloud Virtual Machines (VMs) are the workhorses of modern infrastructure, offering dedicated slices of CPU, RAM, and storage without the overhead of physical server maintenance. By leveraging a hypervisor to partition hardware, these instances allow you to spin up environments like Ubuntu or Windows Server in seconds (often under 60 seconds for standard images). They provide the flexibility to scale resources vertically as workloads grow or horizontally across global regions to ensure low latency. Whether you are running a high-traffic web server, a Jenkins build node, or a complex SQL database, Cloud VMs deliver the precise control of a private server with the elasticity of the cloud.
Provision and manage resizable compute capacity (virtual servers, or 'instances') on the AWS cloud.
Amazon Elastic Compute Cloud (EC2) delivers secure, scalable compute infrastructure: We offer the industry's broadest platform, featuring over 1000 instance types optimized for diverse workloads (e.g., General Purpose, Compute Optimized). You select an Amazon Machine Image (AMI) — essentially a template with your OS and software — to launch a virtual server. EC2 supports multiple operating systems, including Amazon Linux, Ubuntu, Windows Server, and macOS. Key services like Auto Scaling and Elastic Load Balancing (ELB) are integrated to automatically adjust capacity and distribute traffic across instances, guaranteeing high availability and performance.
Compound Engineering is an AI-native development philosophy where every completed task acts as an investment that accelerates all future work.
Traditional engineering follows a linear path: as codebases grow, complexity increases and development slows. Compound Engineering flips this trajectory by treating every bug fix, code review, and architectural decision as a permanent asset in a learning loop. By utilizing AI agents to codify tribal knowledge and automate repetitive patterns, teams achieve 300% to 700% productivity gains. The system follows a tight four-step cycle (Plan, Work, Review, Compound) to ensure that the environment learns from every iteration. This approach allows single operators to manage complex products like Cora and Sparkle with the leverage of an entire traditional engineering department.
OpenClaw is a viral, open-source personal AI assistant that connects LLMs to your messaging apps and local system to autonomously run tasks 24/7.
OpenClaw turns large language models into proactive, self-hosted personal agents that live inside your daily communication channels (like Telegram, Slack, and Discord). By running on your own hardware or a VPS, it securely bridges LLMs with your local files, terminal, and custom APIs to execute real-world tasks (such as auto-generating GitHub pull requests, monitoring servers, and running cron-based workflows). With support for both commercial APIs and fully local models via Ollama or LM Studio, OpenClaw gives you complete, cost-free control over your private AI infrastructure.